AGENT SKILL

evaluate-help

Overview of evaluate commands

License 许可
MIT
所属仓库
cooneycw/skillc
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
evaluate-help
description
Overview of evaluate commands

02 资源文件 4

03 SKILL.md 正文

<!-- GENERATED by claude-power-pack - scripts/codex-skill-sync.py; edit .claude/commands/evaluate/help.md instead -->

Codex harness adaptations

Generated from a Claude Code command. Where the procedure references these Claude-only surfaces, adapt as follows:

  • MCP tools: use the MCP servers configured in ~/.codex/config.toml, or fall back to the referenced repo scripts and CLI entry points.
  • Helper scripts are bundled under scripts/ in this skill directory (byte-identical copies from the claude-power-pack checkout). Commands run them as "${CPP_SKILL_DIR:-/dev/null}"/scripts/<name>: CPP_SKILL_DIR is the absolute directory containing THIS SKILL.md - the location the skills catalog gives, with any alias such as r0 expanded. A shell resolves a bare scripts/<name> against your working directory (the target project), never this skill, so substitute that directory, or set it in the same command (CPP_SKILL_DIR=/abs/skill/dir; ...) - each command is a fresh shell. "Not a directory" on /dev/null/scripts/<name> (exit 126, or 127 in some shells) means CPP_SKILL_DIR is unset: set it and re-run; never run the target project's own scripts/<name> in its place. Read every "exit 127" fallback in this skill as "exit 127, or 126 Not a directory": set the variable and re-run before falling back. A helper whose verdict gates a step and that prints no verdict line after the whole fallback chain did not run - for the finish gate, no FLOW_FINISH_GATE: verdict line: STOP, never read it as skipped, never commit on it (issue #1547). A bare scripts/<name> inside a command is a CPP-checkout operation that the text says so about; in prose, scripts/<name> NAMES the bundled helper. Some helpers expect sibling repo resources, so prefer a full checkout when one is available.
  • Network: this skill's bundled helpers appear to call the network (gh, git fetch/push, curl, aws, HTTP). Codex's workspace sandbox has no network, and an allow rule for gh covers only a top-level gh command, never one a helper runs. Run these helpers with escalated permissions. From inside the sandbox, "error connecting to api.github.com" means no network, not a GitHub outage or a bad login.

Evaluate Commands

Multi-model evaluation flow for structured decision-making and spec generation.

Commands

Command Purpose
/evaluate-issue Full 4-phase evaluation: divergence scan, reasoning, validation, spec output
/evaluate-help This help page

How It Works

The evaluate flow uses multiple LLM models to analyze an issue or idea from different perspectives, then synthesizes the analysis into spec-ready artifacts.

Input (description + domain)
  ↓
Phase 1: Multi-Model Divergence Scan
  → Models analyze independently, surfaces disagreements
  → Human Checkpoint: select focus areas
  ↓
Phase 2: Sequential Reasoning (12-15 steps)
  → Structured problem → options → trade-offs → convergence
  → Human Checkpoint: accept, redirect, or skip
  ↓
Phase 3: Multi-Model Validation
  → Models validate the recommendation against original proposal
  → Human Checkpoint: choose output type
  ↓
Phase 4: Spec Output
  → Generates .specify/specs/{feature}/ (spec.md, plan.md, tasks.md)

Domain Types

Domain Best For
architecture System design, APIs, infrastructure, data models
concept Ideas, workflows, business logic, feature proposals
algorithm Performance, data structures, optimization
ui-design UX flows, components, accessibility
workflow CI/CD, automation, tooling, DX

Prerequisites

  • Required: the Second Opinion MCP server, registered at user scope (external cooneycw/mcp-second-opinion server; see /second-opinion-help for setup)
  • Optional: Sequential Thinking MCP (enhances Phase 2 reasoning; falls back to inline reasoning if not installed)
  • Skill: evaluate skill provides domain-specific prompts (loaded automatically)

Quick Example

# Evaluate an architectural decision
/evaluate-issue "Should we use event sourcing or CRUD for the order system?"

# Evaluate a feature concept
/evaluate-issue "Add real-time collaboration to the document editor"

# No arguments - interactive prompts guide you
/evaluate-issue

Output

Generates spec artifacts in .specify/specs/{feature-name}/:

.specify/specs/{feature-name}/
├── spec.md      ← Requirements, user stories, acceptance criteria
├── plan.md      ← Technical approach, architecture, risks
└── tasks.md     ← Actionable items organized in waves

Use ./scripts/speckit-tasks-to-issues.sh to create GitHub issues from the generated tasks.

Dependency: scripts/speckit-tasks-to-issues.sh calls scripts/speckit-context.py at runtime to render each issue's task-context block (#858). Both ship together: if the helper is missing or fails, the converter stops with an error before creating anything, because a packaging fault is not a context-free installation. --no-context is the explicit opt-out.

Cost

Typical evaluation costs $0.10-0.30 depending on model selection and depth. Phase 1 and 3 use multi-model calls; Phase 2 uses sequential reasoning (no external LLM cost if using Sequential Thinking MCP).

Related Commands

  • /second-opinion-start - Quick single-file code review
  • /second-opinion-models - Interactive model selection for reviews
  • /spec:adopt - Install the official spec-kit for structured spec authoring
  • scripts/speckit-tasks-to-issues.sh - Turn a spec tasks.md into GitHub issues

04 同仓库更多 Skills